Unbrowser is the SSR/static-first tier for public web tasks, with optional bounded QuickJS and no rendering. Playwright MCP provides full-browser automation with persistent, isolated, and extension-connected state. Choose Unchained MCP for its DDM-oriented local bridge and agent-client setup.
Unbrowser is not a drop-in replacement for every Playwright workflow. Browser profile and login state are not exclusive to Unchained: Playwright MCP supports persistent and isolated profiles plus an extension path to existing Chrome tabs. Choose by runtime and tool surface.
Use a stateful native runtime for SSR/static-first public pages when pixels are unnecessary. Default navigation skips scripts; optional bounded QuickJS can execute some page scripts, while heavy hydration may remain partial.
Use a full browser when the job depends on rendered visual state, screenshots, layout, canvas, WebGL, or Playwright's testing and automation surface.
Use Unchained when its DDM page-layout output, CDP tools, local agent/bridge, and setup paths for Claude, Codex, or OpenCode fit the workflow. It is not the only path to logged-in browser state.
Direct MCP tool calls invoke browser actions directly; automatic confirmation is not implied. Hosted Agent View has a separate confirmation and review flow.
Set up Unchained MCP →Adjust the workload inputs to explore one planning scenario. The runtime, memory, cold-start, token, duration, and infrastructure values are fixed assumptions in this page—not universal benchmarks, provider quotes, or guaranteed savings.
Presets are illustrative. Real outcomes vary by tool and browser version, page shape, output format, concurrency, cache state, host, and pricing. Validate the same task on your own infrastructure before choosing capacity or estimating cost.
These outputs extend the declared workload and cost assumptions linearly. They are planning comparisons, not measured bills, quotes, or guaranteed savings.
A lighter tool is useful only when it handles the task. This high-level matrix is a routing aid, not a compatibility guarantee. Verify edge cases against each project's current documentation.
| Capability | curl | WebFetch | unbrowser | Playwright | Browserbase |
|---|
This page multiplies the per-session assumptions linearly from 1 to 500 sessions. Real systems may share processes, queue work, cache state, or hit nonlinear limits.
Use this as a first routing question, then verify the chosen tool on the actual page and workflow.
The table below declares the page's per-tool binary, memory, cold-start, and token inputs. Total memory is per-session memory multiplied by concurrent sessions and a deployment multiplier. Token volume is tokens per page multiplied by pages per task and tasks per day. The compute model assumes 0.05 hours per task and applies the page's illustrative infrastructure rate; token cost uses the rate selected above. None of those constants adapts to a real runtime, provider bill, cache, page, or model.
Reproduce the same workload on the versions and infrastructure you plan to use before treating a difference as operational or financial evidence.
These constants drive the calculator. They are visible so you can judge the model, not because they are universal values for every workload.
Try the fixed public-site Unbrowser demo without an account, or set up Unchained MCP when you want its DDM tools and local bridge. Playwright MCP remains a full-browser option for pixels and existing browser state.